event, p2p/simulations/adapters: use buffered channel to avoid goroutine leak

[Chain-Fox]

subscribe() in event/subscribe.go and Stop() in p2p/simulations/adapters/exec.go share a same problem: some goroutines in these functions will be leaked in certain cases (unsub for subscribe(), and timeout for Stop()).
In subscribe(), the send operation subscribed <- err on L115 will be blocked forever when unsub is selected. Thus the child goroutine leaks and always occupies the memory.
Adding one buffer to the channel will make the send operation never block the child goroutine,
and it will not change semantic.
The same is true for Stop().
In Stop(), although Kill() is called on timeout,
I am not sure if the child goroutine will be in the same process or not. Here is the annotation of Kill(), in /os/exec.go:

// Kill causes the Process to exit immediately. Kill does not wait until
// the Process has actually exited. This only kills the Process itself,
// not any other processes it may have started.
func (p *Process) Kill() error {
    return p.kill()
}

[fjl]

Very nice finding!

[fjl]

I think it would be better to fix this case by waiting on the subscribed channel when the subscribe loop exits. This way, Unsubscribe will block until the goroutine running s.fn has exited.

[fjl]

Actually, I have now applied the fix myself.

[fjl]

This fix is OK as-is.